llama.cpp

metal : disable fast math in all quantize kernels

#14528

Merged